The paradigm of edge computing is pivotal for deploying deep learning object detectors in time-sensitive applications. Nevertheless, practical efficacy is often impeded by critical impediments: 1) the inherent trade-off between detection precision and model lightweightness; 2) the inflexibility of generalized deployment frameworks for task-specific object detection; and 3) the scarcity of validation in real world operational environments. To address these challenges, we propose the Edge Detection Toolbox (ED-TOOLBOX), which leverages generalizable plug-and-play components to enable edge-site adaptation of object detection models. Specifically, we propose a lightweight Reparameterized Dynamic Convolutional Network (Rep-DConvNet) that employs a weighted multi-shape convolutional branch structure to enhance detection performance. Furthermore, ED-TOOLBOX includes a Sparse Cross-Attention (SC-A) network that adopts a localized-mapping-assisted self-attention mechanism to facilitate a well-crafted Joint Module in adaptively transferring features for further performance improvement. Moreover, we propose an Efficient Head for the classification and location modules to achieve more efficient prediction. Furthermore, we address a critical oversight in industrial safety: conventional helmet detection's neglect of band fastening. To bridge this gap, we construct the Helmet Band Detection Dataset (HBDD) and deploy our ED-TOOLBOX-optimized model on this practical challenge. Extensive experiments validate the efficacy of our components. In surveillance simulations, our model surpasses six state-of-the-art methods, achieving both real-time performance and high accuracy. These results establish our approach as a superior solution for edge object detection.

Federated Learning (FL) enables collaborative model training across decentralized clients without sharing private data. However, FL suffers from biased global models due to non-IID and long-tail data distributions. We propose \textbf{FedSM}, a novel client-centric framework that mitigates this bias through semantics-guided feature mixup and lightweight classifier retraining. FedSM uses a pretrained image-text-aligned model to compute category-level semantic relevance, guiding the category selection of local features to mix-up with global prototypes to generate class-consistent pseudo-features. These features correct classifier bias, especially when data are heavily skewed. To address the concern of potential domain shift between the pretrained model and the data, we propose probabilistic category selection, enhancing feature diversity to effectively mitigate biases. All computations are performed locally, requiring minimal server overhead. Extensive experiments on long-tail datasets with various imbalanced levels demonstrate that FedSM consistently outperforms state-of-the-art methods in accuracy, with high robustness to domain shift and computational efficiency.

Large language models (LLMs) (e.g., ChatGPT, GPT-4 and Sora) have fundamentally transformed our daily lives, catalyzing breakthroughs in natural language processing, computer vision and revolutionizing human-computer interactions. However, their transformative potential is hindered by immense computational resources required. To address this challenge, the integration of LLMs with edge-cloud computing has become a focal point in advancing the capabilities of artificial intelligence. This survey comprehensively exploits the landscape of edge-enhanced intelligence, specifically focusing on the synergy between LLMs and edge-cloud computing. We delve into the evolution of LLMs, architectural intricacies, and the computational challenges associated with deploying them in edge environments from the aspects of data, computing power, model training and inference. Furthermore, we exploit the bidirectional symbiotic relationship between edge-cloud computing and LLMs from two key aspects: edge-cloud computing empowered LLMs (Edge4LLMs) and LLMs driven edge-cloud computing (LLMs4Edge). After that, we examine the dynamic collaboration between edge-cloud computing and LLMs, highlighting the complementary roles in optimizing the efficiency, real-time, and scalability of intelligent applications. Through an extensive review of existing research and practical implementations, this survey offers insights into the current state of edge-enhanced intelligence, identifies key challenges, and proposes potential avenues for future research and development. This survey aims to provide a comprehensive understanding of the emerging paradigm of edge-enhanced intelligence for LLMs, thereby fostering informed discussions and inspiring advancements in this field.

With the growing intelligence of Internet of Vehicles (IoV) applications, high-real-time tasks impose stricter requirements on information freshness and computational efficiency. To address these requirements, computation offloading alleviates computational resource bottlenecks on vehicular devices by offloading computing tasks to edge servers. Moreover, the frequent information exchanges involved in the offloading process significantly increase the risk of exposing sensitive vehicle location data and lead to serious location privacy leakage. More critically, attackers can further infer sensitive information such as vehicle owner identities, behavioral patterns, and social relationships from leaked location data. To address the above challenges, this paper proposes a computation offloading framework that integrates Age of Information (AoI) with Geo-Indistinguishability (Geo-I). We firstly introduce an AoI model to quantify information freshness, then employ a Geo-I mechanism to perturb actual vehicle locations, preventing privacy leakage. In addition, we design a dynamic privacy budget adjustment method to adaptively balance privacy protection strength and data availability. To handle complex decision-making in dynamic IoV environments, we learn offloading strategies based on the Group Relative Policy Optimization (GRPO) algorithm to maximize the long-term system utility. Experiments on real-world datasets demonstrate that the proposed approach outperforms baseline algorithms and achieves an effective balance between privacy protection and data utility under the tested settings.

In the mobile edge computing (MEC)-enabled Internet of Vehicle (IoV) networks, the communication quality between vehicles and edge servers during task computation offloading is vulnerable to degradation caused by obstacle interference, thereby reducing user quality of service. The Intelligent Reflecting Surface (IRS) is a passive and programmable device, can be installed on the exterior of buildings to reflect and enhance signals through phase shift adjustment. However, a coupling relationship exists between IRS phase control and task offloading decisions. Optimizing them separately by decoupling fails to reflect the highly coupled and dynamic interactions in real systems. This may lead to a mismatch between offloading decisions and channel conditions, making it difficult to achieve optimal system performance in real-world environments. To address this challenge, we propose a joint optimization method for IRS phase control and task computation offloading based on causal inference (OITC). Specifically, we first analyze and model the dynamic channel between vehicles and the IRS. Then, causal inference is used to verify the causal relationship between IRS phase control and task computation offloading. Moreover, we combine causal inference with the Multi-agent Proximal Policy Optimization (MAPPO) algorithm to jointly optimize IRS phase control and task offloading decisions. Ultimately, extensive experiments demonstrate that the proposed algorithm outperforms traditional multi-agent deep reinforcement learning (MADRL) methods in terms of communication quality, system delay, and energy consumption.

In agentic systems, human-generated data records anchor the value of AI services. Yet cloud compute pipelines centralize processing on remote servers. Data centralization reduces personal data sovereignty and may potentially degrade the quality of service (QoS). Meanwhile, user contributions are diverse in quantity and quality: decentralized records can be biased, noisy, and heterogeneously distributed. To address the data challenge, we study fair token allocation and private data valuation for decentralized and resource-constrained agentic systems. Our approach embeds multi-modal representations in a shared semantic space and releases differentially private (DP) prototypes to preserve utility while reducing semantic leakage. With the DP guarantee, we design a fair token allocation scheme that rewards effective contributions and remains robust to data heterogeneity and AI resource scarcity. Extensive simulations demonstrate improved contribution-based fairness and QoS compared to standard benchmarks. The improved resistance to image reconstruction attacks indicates enhanced privacy for multi-modal personal data.

In large-scale Industrial Internet of Things (IIoT) scenarios such as smart ports and open-pit mines, wide-area visual perception is essential for intelligent monitoring, and it may serve as a perception basis for future digital twin applications. Bionic compound-eye cameras provide an ultra-wide field of view and hundred-megapixel resolution, enabling single-node panoramic sensing in Wireless Visual Sensor Networks (WVSNs). However, compound-eye imagery also introduces two major difficulties: heavy transmission overhead and weak perception of tiny targets caused by extreme object-scale variation. To address these issues, this paper proposes an edge-cloud collaborative Internet of Video Things (IoVT) perception framework. In this framework, high-resolution visual streams are processed directly at edge nodes, and only high-value semantic information is transmitted to the cloud. As the core edge inference engine, we develop DPGL-YOLO, a lightweight detector tailored to compound-eye imaging characteristics and edge deployment constraints. Specifically, the proposed method introduces a Dual-Path Attention Fusion Module (DPAM) to enhance cross-scale feature interaction, a Global–Local Collaborative Attention Mechanism (GLAM) to enhance global-local discrimination under cluttered backgrounds, a high-resolution P2 detection head, and an MIF-WiseIoU regression loss to improve localization of dense small objects under weak-feature and cluttered-background conditions. Experiments on two in-house hundred-megapixel compound-eye datasets collected from real port and mining environments show that DPGL-YOLO achieves 86.84% mAP@0.5 and 60.77% mAP@0.5:0.95 on the port dataset, and 72.30% mAP@0.5 and 30.68% mAP@0.5:0.95 on the mining dataset. The proposed method outperforms representative detectors while maintaining edge-oriented inference efficiency. The proposed framework provides an edge-oriented perception architecture that may help reduce transmission burden in industrial IoVT systems.

This paper establishes Unmanned aerial vehicles (UAVs)-assisted integrated sensing and communication (ISAC) networks in the presence of high-speed mobile eavesdropper UAV (EUAV). The source UAV (SUAV) provides communication services for ground users (GUs), while performing real-time sensing of the EUAV's position. Leveraging the intelligence provided by the SUAV, the jamming UAV (JUAV) dynamically tracks the EUAV and projects directional jamming beamforming. To maximize the network's secrecy rate, this paper proposes a four-tier joint iterative optimization algorithm, which optimizes the SUAV's communication scheduling, the allocation of communication and sensing power, as well as the motion parameters of both the SUAV and JUAV. Simulation results demonstrate that the proposed scheme significantly improves the network's secrecy rate.

Semantic communications (SemCom) is a promising paradigm that prioritizes the transmission of task-relevant information, thereby enabling superior communication efficiency over traditional bit-centric systems. However, most existing SemCom systems face critical limitations in computational efficiency and spatial flexibility. To overcome these limitations, we propose a novel unmanned aerial vehicles (UAV)-enabled distributed electromagnetic neural network (EMNN) for a task-oriented SemCom system. Specifically, the proposed distributed EMNN is composed of multiple UAV-mounted stacked intelligent metasurfaces (SIM) and a ground receiving station (GRS), where multiple SIMs collaboratively encode image semantics in the wave domain, and the GRS performs decoding based on the received power distribution. Moreover, we employ a temperature-adaptive gradient optimization algorithm to train the distributed EMNN, which mitigates gradient vanishing and enhances learning stability. Finally, the numerical simulation results demonstrate the effectiveness of distributed EMNN in image recognition task-oriented SemCom, achieving an average 8% accuracy improvement over the single-SIM baseline across multiple datasets.

Edge Intelligence (EI) combines edge computing and artificial intelligence to deliver low-latency and resource-efficient services. Integrated Sensing and Communication (ISAC) further empowers EI by enhancing edge perception and accelerating intelligent model training. However, integrating ISAC into EI complicates the coordination of dynamically varying sensing, communication, and computation resources, especially under device mobility and unpredictable network conditions, leading to degraded service performance. To address these coordination challenges and sustain high-quality service under mobility and dynamics, we aim to design an adaptive service provision framework that tightly couples real-time perception with intelligent decision-making at the edge. Specifically, we propose an adaptive service provision architecture for ISAC-assisted EI, where Digital Twins (DTs) hosted on edge servers represent edge devices and their contexts to enable accurate perception and intelligent decision-making, thereby enhancing the efficiency of ISAC-enabled services. By dynamically migrating DTs across edge servers based on device mobility and resource availability, the system supports continuous decision-making and seamless service delivery. We further integrate convex optimization for efficient multi-resource coordination and a Time-Varying Contextual Bandit (TVCB) algorithm to enable adaptive, context-aware DT migration in dynamic environments. Extensive simulations demonstrate that our approach significantly improves service quality, reliability, and adaptability in ISAC-assisted EI systems, reducing migration oscillations and overhead while achieving lower latency and higher utility compared with representative baselines.

Since the release of GPT2-1.5B in 2019, the large language models (LLMs) have evolved from specialized deep models to versatile foundation models. While demonstrating remarkable zero-shot ability, the LLMs still require fine-tuning on local datasets and substantial memory for deployment over the network edges. Traditional first-order fine-tuning techniques require significant GPU memory that exceeds the capacity of mainstream hardware. Besides, the LLMs have been expanded beyond text generation to create images, audio, video, and multi-modal content, necessitating careful investigation of efficient deployment strategies for large-scale foundation models. In response to these challenges, model fine-tuning and model-compression techniques have been developed to support the sustainable growth of LLMs by reducing both operational and capital expenditures. In this work, we provide a comprehensive overview of prevalent memory-efficient fine-tuning methods for deployment at the network edge. We also review state-of-the-art literature on model compression, offering insights into the deployment of LLMs at network edges.

Band selection (BS) is a key method in hyperspectral image (HSI) classification that helps to reduce the computational burden and improve the class separability. However, with the emerging of uncrewed aerial vehicle (UAV)-borne HSI datasets, their attributes, such as high spatial and spectral resolution as well as large-scale samples pose serious challenges to the existing BS methods, making them inefficient. In addition, the efficient utilization of the prior knowledge from the data collected by fixed UAV-borne sensors in different regions is often easily overlooked. In view of these issues, this article proposes a neural network-assisted particle swarm optimization (PSO) algorithm for cross-domain BS of UAV-borne HSIs. First, a knowledge learning strategy is designed for the source domain, which applies a neural network model to learn the useful prior knowledge in labeled source domain data. Then, a network-assisted PSO algorithm is introduced to search for the optimal subset of bands in the target domain under the guidance of the valid prior knowledge captured from the source domain by the network model. Moreover, a similarity-based grouping strategy is designed to group similar bands and then select bands from each group with the aims of reducing the redundant information in the subset of bands. Finally, experimental results on three common UAV-borne HSI datasets show that our proposed method can efficiently handle UAV-borne HSI data with large samples, as it is able to find a subset of bands with higher quality compared to several state-of-the-art BS methods.

Wireless Time-Sensitive Networking (WTSN) represents a cutting-edge technological advancement designed to meet the stringent requirements of time-sensitive applications. A fundamental challenge in WTSN implementation lies in achieving efficient traffic scheduling and routing that simultaneously guarantees good scalability, high success rate, and low delay. To overcome these challenges, this paper presents Trident, a tri- fold approach integrating Frame Dynamic Priority Scheduling (FDPS), Level of Collision (LoC)-aware traffic grouping (LATG), and LATG-aware multipath selection and routing (LAMSR). Specifically, FDPS dynamically adjusts frame priorities based on real-time attributes of deadlines, inter-arrival time, frame length, and hop counts, while LATG leverages graph theory to partition traffic flows with minimal collision risks. LAMSR optimizes multi-path routing using real-time metrics obtained form LATG and FDPS to ensure minimal delays. Key findings in experiments demonstrate that Trident achieves better transmission success rates under large number of flows, maintains stable scheduling latency even at elevated uplink power levels, and sustains enhanced scalability in multi-hop scenarios, outperforming the baseline methods.

Emerging network paradigms and applications increasingly rely on federated learning (FL) to enable collaborative intelligence while preserving privacy. However, the sustainability of such collaborative environments hinges on a fair and stable payoff allocation mechanism. Focusing on coalition stability, this paper introduces a payoff allocation framework based on the least core (LC) concept. Unlike traditional methods, the LC prioritizes the cohesion of the federation by minimizing the maximum dissatisfaction among all potential subgroups, ensuring that no participant has an incentive to break away. To adapt this game-theoretic concept to practical, large-scale networks, we propose a streamlined implementation with a stack-based pruning algorithm, effectively balancing computational efficiency with allocation precision. Case studies in federated intrusion detection demonstrate that our mechanism correctly identifies pivotal contributors and strategic alliances. The results confirm that the practical LC framework promotes stable collaboration and fosters a sustainable FL ecosystem.

As a promising paradigm for next-generation wireless networks, semantic communication has garnered widespread attention due to its significant potential to enhance the transmission efficiency of various tasks. However, existing multi-user semantic communication systems primarily rely on orthogonal multiple access (OMA), which leads to inefficient spectrum utilization in resource-constrained environments. To further enhance utilization of resources across different domains, this paper proposes a novel hybrid non-orthogonal multiple access (NOMA)-assisted multi-task semantic communication networks. Specifically, users in the OMA-based legacy network are allowed to access multiple orthogonal subcarriers, aiming to enhance the transmission flexibility and bandwidth utilization. Quality of experience (QoE) is employed as a measure of performance in our work to facilitate the optimization of radio resources. By jointly optimizing the transmit power decision scheme and the average transmitted semantic symbols per user, the issue is formulated to maximize the overall effective QoE for multiple users. Considering the non-convex characteristics of dynamic wireless environments, a resource allocation framework utilizing deep reinforcement learning (DRL) is designed. Simulation results demonstrate that the hybrid NOMA-based multi-user semantic communication framework outperforms the OMA-based benchmark, which highlights the superior performance of the proposed framework. Moreover, the proposed DRL-based algorithm significantly improves the total QoE of the multi-task semantic communication networks, which reflects the superiority of the designed optimization scheme.

To address the challenges of information leakage, low energy efficiency, and the Doppler effect in mobile Internet of Vehicles (IoV), this paper proposes an enhanced IoV cooperation framework, where privacy information is forwarded by the untrusted relay assisted by uncrewed aerial vehicles (UAVs) and reconfigurable intelligent surface (RIS), which can improve security and energy efficiency. To meet the requirements of green communication, we formulate a secrecy energy efficiency maximization problem by jointly optimizing the transmit power allocation, the relay's amplification factor, the two-hop RIS phase shift matrices, and the UAV trajectory. Given the non-convex nature of this problem, we introduce an iterative algorithm based on the convex-concave procedure and Dinkelbach's method to optimize the transmit power and amplification factor. Additionally, we conceive the majorization-minimization (MM) algorithm to optimize the two-hop RIS phase shift matrices, and a designed firefly algorithm-deep deterministic policy gradient (FA-DDPG) algorithm is proposed to obtain the UAV trajectory. Simulation results demonstrate the effectiveness of the proposed scheme in enhancing secrecy energy efficiency. Specifically, compared to the DDPG-only and FA-based schemes, the proposed scheme achieves an improvement of 33.3% and 64.2%, respectively, in secrecy energy efficiency.

This comprehensive survey paper examines the applications of artificial intelligence (AI) in unmanned aerial vehicle (UAV)-enabled wireless networks. With the increasing demand for efficient and adaptive communication systems, the integration of AI with UAV networks promises to revolutionize various aspects of wireless communication. The paper first outlines the background and motivation behind AI integration, highlighting the potential for enhanced network performance, autonomy, and adaptability. It then delves into the key AI applications across different network layers, including data sensing and collection, placement and trajectory optimization, radio resource management, routing and topology control, edge computing and caching, as well as security and privacy enhancement. For each application, the paper discusses relevant AI techniques, main findings, optimization objects, and the potential benefits and challenges. The survey also identifies open issues, such as the practical implementation gap, standardization issues, and real-world application barriers, and proposes future directions to address these challenges and further advance the field. In conclusion, the integration of AI with UAV-enabled wireless networks (UWNs) holds tremendous potential for transforming wireless communication, enabling new applications and services with unprecedented capabilities.

Stacked intelligent metasurfaces (SIM) provide a promising physical architecture for efficient deployment of semantic inference in wireless systems. However, existing SIMs in practical semantic communication (SemCom) systems are limited to phase-only control, yielding limited frequency-domain degrees of freedom and insufficient capacity for hierarchical semantic encoding under constrained parameter and energy budgets. To overcome this, we propose a harmonic electromagnetic neural network (EMNN) built on SIMs, which introduces multi-harmonic modulation at each meta-atom to expand frequency-domain degrees of freedom beyond phase-only control. In addition, a multi-scale sliding-window source-encoding scheme is introduced to improve robustness against spatial variations of the system. The following layers perform frequency mixing for semantic encoding, progressively generating higher-order harmonics for additional degrees of freedom. Importantly, the multi-harmonic components are further fused into the center harmonic, producing an effective nonlinear representation gain that strengthens feature separability. To support passive implementation and improve calibration, passivity-oriented normalization and a temperature-like scaling are further adopted. The simulation results demonstrate that the proposed method outperforms harmonic-free baselines by 10% in classification accuracy.

With the rapid development of 6G and Internet of Vehicles (IoV) technologies, the volume of computation-intensive tasks generated by intelligent vehicles is growing exponentially. Given limited onboard processing capabilities, vehicles increasingly rely on edge servers deployed by service providers (SPs) at roadside units to offload tasks. Vehicle clients can offload the tasks to SPs to mitigate their onboard computation load, while SPs derive economic benefits through the provision of computation resources. However, this interaction introduces a conflict of interest, as vehicles aim to minimize their offloading costs, while SPs seek to maximize revenue. To address this problem, we propose SPOR, a Stackelberg game-based service priority-aware computation offloading and resource pricing scheme in IoV. SPOR is a hierarchical game-theoretic framework in which SPs act as leaders setting prices, while vehicles act as followers determining their offloading strategies. A novel service prioritization function is introduced, incorporating booking price, system load, and reputation to ensure fair and balanced resource allocation. We provide a theoretical proof of the existence and uniqueness of a Nash equilibrium. Extensive experiments on a real-world vehicle edge computing dataset show that SPOR outperforms baseline methods in delay, energy consumption, average load, and task completion rate. Notably, SPOR maintains task completion rates above 97% even under heavy workloads, demonstrating its effectiveness in enhancing system reliability and overall performance.

Large language models (LLMs) have demonstrated exceptional proficiency in language understanding. However, when LLMs align their outputs with deceptive and/or misleading prompts, the generated responses could deviate from the de facto information. Such observations are known as fawning hallucinations, where the model prioritizes alignment with the input's implied perspective over accuracy and truthfulness. In this work, we analyze fawning hallucinations in various natural language processing tasks and tailor the so-termed contrastive decoding method for fawning-hallucination mitigation. Specifically, we design two paradigms to generate corresponding deceptive and/or misleading inputs for the consistent fawning hallucinations induction. Then, we propose the collaborative contrastive decoding (CCD) to handle the fawning hallucinations across different tasks in LLMs. By contrasting the deviation in output distribution between induced and transformed neutral inputs, the proposed CCD can reduce reliance on deceptive and/or misleading information without requiring additional training. Extensive experiments demonstrate that the proposed CCD can effectively mitigate fawning hallucinations and improve the factuality of the generated responses over various tasks.